1. Data Cleaning and Wrangling¶
Lets take a look at our data:¶
zomato.info()
zomato.head()
After a quick look at the columns of the dataframe, I have decided to remove some columns which will not be very useful in the Data Analysis process like url, address, phone, menu_item and location as they are redundant.¶
After slicing off the unwanted columns, below is the final list of columns:
# zomato.columns
set(zomato)
Now lets make the rate column more appropriate for Analysis by converting it to float type and removing '/5' from the rating"¶
Also Renaming our column names to more appropriate names:¶
zomato.rename(columns={'cuisines' : 'Cuisine', 'listed_in(city)': 'Locality', 'listed_in(type)' : 'Listed_Type', 'approx_cost(for two people)': 'Approx_Cost', 'name': 'Name', 'rest_type':'Restaurant_Type', 'rate' : 'Rating', 'votes' : 'Total_Votes', 'online_order': 'Online_Order','book_table':'Table_Booking', 'dish_liked' : 'Dishes_Liked' }, inplace = True)
zomato.columns
Dropping all Nans, Nulls and non-numeric characters from the dataset:¶
# Lets drop the data points with na and 0 rates:
zomato.drop(zomato[(zomato['review_rates'] == 0) & (zomato['Rating'].isna())].index, inplace=True)
# Replacing NaNs and '-' of rate with review rate values:
zomato['Rating'] = np.where(zomato.Rating.isna(), zomato.review_rates, zomato.Rating)
zomato['Rating'] = np.where(zomato['Rating'] == '-', zomato.review_rates, zomato.Rating)
#After dropping Nas
zomato.head()
# We have few columns with NaNs so lets drop these
print("Approx_Cost column has {} Nan values and Cuisine column has {} Nan values, we need to drop these.".format(sum(zomato.Approx_Cost.isna()), sum(zomato.Cuisine.isna())))
Now lets convert Rating, Approx_Cost to numeric data type and Remove the non- ascii characters from Name column.¶
# print(type(zomato['Approx_Cost'][0]))
zomato['Rating'] = zomato['Rating'].apply(pd.to_numeric)
zomato['Approx_Cost'].replace(',', '',regex=True, inplace=True)
zomato['Approx_Cost'] = zomato['Approx_Cost'].apply(pd.to_numeric)
zomato['Name'].replace({r'[^\x00-\x7F]+':''}, regex = True, inplace=True)
# type(zomato['Approx_Cost'][0])
zomato.head()
So now we can say that our DataFrame is clean and ready for Analysis. Lets take a quick look at our data info:¶
zomato.info()
2. EDA - Visualization¶
Q) How many restaurants provide option of ordering online?¶
fig1, ax1 = plt.subplots(figsize=(6,5));
ax1.pie(zomato.Online_Order.value_counts().values, autopct = '%1.1f%%', shadow=True);
ax1.legend(zomato.Online_Order.value_counts().index,
loc="lower right")
ax1.axis('equal');
ax1.set_title('Number of restaunrants with online ordering');
We see that almost 65% of the restaurants have the online ordering facility, as it is one the major source of income
Q) How many restaurants provide table booking, how are they different from the ones that dont?¶
zomato.Table_Booking.value_counts().plot.bar();
plt.title('Number of restaurants with online table booking');
plt.xticks(rotation = 0);
So here we see that most of the restaurants don't provide table booking option. They are very few, only around 5k out of almost 50k total so Lets look at the kind of restaurants these are.
plt.figure(figsize=(10,6))
ax = zomato[zomato['Table_Booking'] == 'Yes'].Restaurant_Type.value_counts(ascending = True).tail(20).plot.barh(y = 'Restaurant_Type');
plt.xlabel('Number of Restaurants')
plt.title('Distribution of Types of Restaurants with Table Booking option');
ylab = zomato[zomato['Table_Booking'] == 'Yes'].Restaurant_Type.value_counts(ascending = True).tail(20).values
for i, v in enumerate(ylab):
ax.text(v+50, i-0.25, np.around(v/sum(ylab)*100,2), color='black', fontsize = 10)
We can see from the graph that More than 55% of the restaurants with Table Booking option available are Casual Dining type and 9% are Cafes.
Lets check the average rating of Restaurants with and without Table Booking facility:
print(zomato[zomato['Table_Booking'] == 'No'].Rating.describe())
print("The average rating of the restaurants without table booking option is 3.6 with 50% higher than 3.7/5. Also size of this dataset is larger. Is the cost also lower at such restaurants?")
print(zomato[zomato['Table_Booking'] == 'Yes'].Rating.describe())
print("There are 6k restaurants with table booking facility and have an average rating of 4.1 with 50% of them having more than 4.3/5 rating which shows that restaurants with table booking receive better rating. Is it truly the case?")
print("The minimum cost at restaurants with table booking is Rs.{} while the minimum cost at restaurants without table booking is Rs.{}".format(zomato[zomato['Table_Booking'] == 'Yes'].Approx_Cost.min(), zomato[zomato['Table_Booking'] == 'No'].Approx_Cost.min()))
Hypothesis Testing (T-Test for comparing means of two groups.)¶
Do Restaurants with Table Booking actually receive higher rating? Lets test this hypothesis.
H0 : The mean rating of restaurants with table booking facility is same as mean rating of restaurants without the facility of booking table.
H1 : The mean rating of restaurants with table booking facility is geater than mean rating of restaurants without the facility of booking table.
H$_{0}$: $\mu_{1} == \mu_{2}$
H$_{1}$: $\mu_{1} > \mu_{2}$
Conditions for Hypothesis Testing:
- 1) Samples Are Random
- 2) It is less than 10% of the population
- 3) It can be considered normal as the size is larger than 30.print('Statistics of a random sample of restaurants without table booking facility:')
rand_Rate_Booking_No.describe()
print('Statistics of a random sample of restaurants with table booking facility:')
rand_Rate_Booking_Yes.describe()
#However we will use the function to calculate out p - value:
t_stat, p = ttest_ind(rand_Rate_Booking_Yes,rand_Rate_Booking_No)
print('After running the T-test we get that:')
print('test statistic is', t_stat)
print('and p-Value is', p)
if (p < 0.05):
print("Since P-value is lesser than significance level of 0.05, we reject the Null Hypothesis and Hence there is enough evidence that The mean rating of restaurants with table booking facility is geater than mean rating of restaurants without the facility of booking table. However we can not claim causal as there are other lurking variables.")
else:
print("Since p-value is greater than significance level of 0.05, we fail to reject the Null Hypothesis and hence there is enough evidence that the mean rating of table booking retaurants is same as mean rating of restaurants without the facility of booking table.")
Lets now take a final look at the rate distributions of restaurants with and without table booking:¶
plt.figure(figsize=(30,10))
plt.subplot(1,2,1)
sns.countplot(zomato[zomato['Table_Booking'] == 'Yes']['Approx_Cost'])
plt.xticks(rotation = 60)
plt.title('Rate distribution of restaurants with Booking table option');
plt.subplot(1,2,2)
# plt.figure(figsize=(20,5))
sns.countplot(zomato[zomato['Table_Booking'] == 'No']['Approx_Cost'])
plt.xticks(rotation = 60)
plt.title('Rate distribution of restaurants without Booking table option');
After analyzing further we see that restaurants with Table booking option have higher rating on average (From the Hypothesis) and are costlier than restaurants without booking table option. It start at 300 minimum where as the other type starts at Rs. 40.
From the above graph we see that there are fewer and fewer restaurants with rates above Rs.3000. Lets see what kind of restaurants these are?
Q) What are kind of Restaurants with higher meal cost?¶
print(zomato[zomato['Approx_Cost'] > 3000]['Restaurant_Type'].value_counts())
print('Restaurants with cost higher than 3000 are under Fine Dining. Average cost the restaunrant with Fine Dining is', zomato[zomato['Approx_Cost'] > 3000].Approx_Cost.mean())
plt.figure(figsize=(8,5))
zomato[zomato['Approx_Cost'] > 3000]['Name'].value_counts(ascending=True).plot.barh();
From the above analysis and graph, as expected we see that Fine Dining Restaurants are the costliest type and the most popular chain around bangaluru are Ritz Carlton, JW Marriot, Leela Palace and ITC.
Q) What are the most popular restaurants type around bangalore?¶
plt.figure(figsize=(10, 6))
zomato.Restaurant_Type.value_counts(ascending=True).tail(20).plot.barh();
plt.title('Distribution of Restaurant types:');
From the graph above we see that Quick Bites is the dominating restaurant type in the market, followed by Casual Dining and Cafes. Lets further analyze Quick Bites to find the leading restaurant/ chains.
z1 = zomato[zomato['Restaurant_Type'] == 'Quick Bites']['Name'].value_counts().head(20).reset_index().rename(columns = {'index': 'Restaunrant_Name', 'Name':'Count'})
plt.figure(figsize=(10,10))
sns.barplot(y= 'Restaunrant_Name', x = 'Count', data = z1);
Suprisingly 5-Star chicken, a local brand has over taken McDonalds, KFC and Dominos Pizza around bangalore.
Q) What are the various restaurants types with respect to the type of services they provide?¶
trace = go.Bar(x = zomato.Listed_Type.value_counts().keys(),
y = zomato.Listed_Type.value_counts(),
text = zomato.Listed_Type)
data = [trace]
layout = go.Layout(title = 'Distribution of Restaurants by Service Type', yaxis = dict(title = 'Number of Restaurants'))
fig = go.Figure(data = data, layout= layout)
py.iplot(fig);
Delivery is most popular and adopted by most of the restaurants restaurants!
plt.figure(figsize = (15,10));
# cmap = sns.cubehelix_palette(dark=.3, light=.8, as_cmap=True)
sns.scatterplot(x='Name', y='Approx_Cost', s=40, data = zomato, hue = 'Listed_Type', style = 'Listed_Type', palette='Set2');
plt.xticks('Off')
plt.title('Distribution of Approx_Cost across all Restaurants', size= 18);
Dine-out is more spread out from low cost to high cost meals restaurants.
c1 = ' '.join([text for text in z])
plt.figure(figsize=(10,10))
wordcloud = WordCloud(background_color = 'white', collocations = False, width=1500, height=1500).generate(c1)
plt.imshow(wordcloud)
plt.axis('off')
plt.title('Most popular Dishes in Bengaluru', size = 20)
plt.show()
c2 = ' '.join([st for st in zomato['Cuisine']])
plt.figure(figsize=(10,10))
wordcloud = WordCloud(background_color = 'white', collocations = True, width=1500, height=1500).generate(c2)
plt.imshow(wordcloud)
plt.axis('off')
plt.title('Most popular Cuisines in Bengaluru', size = 20)
plt.show()
Q) What are the most popular places for restaurants in bangalore?¶
plt.figure(figsize=(10,10))
zomato.Locality.value_counts(ascending=True).plot.barh();
Here we see that the locality with maximum restaurants is BTM with almost 3000 restaurants and Koramangala and least is New BEL Road.
plt.figure(figsize=(10,5))
# zomato.Approx_Cost.plot.hist()
sns.distplot(zomato.Approx_Cost)
# plt.xticks(np.arange(0, 4500, 500))
plt.xlim(0, 6000)
plt.xlabel('Approx. Cost for two')
plt.title("The Cost distribution");
We notice that the plot is right skewed with maximum cost being in range 0 to 1000 and very few going beyong 2000. What I would like to investigate further is what kind of restaurants cost higher than 2000 and how is their rating compared to other restaurants.
Lets check the distribution of rating of the restaurants.
plt.figure(figsize=(10,5))
# zomato.Rating.plot.hist();
sns.distplot(zomato.Rating);
Rating plot is left skewed with most of ratings between 3 to 4.5 and maximum at 4. Lets investigate further to see what kind of restaurants receive lower ratings and which restaurants have ratings higher than average.
Lets take a look at the restaurants with maximum number of branches in the city and are most popular.
plt.figure(figsize=(10,6));
zomato.Name.value_counts(ascending = True).tail(20).plot.barh();
CCD seems to have the largest number of chains in Bengaluru, followed by Onesta, Just Bake and Empire. Lets see what factors contribute to its success and popularity. Does CCD , Onesta have comparatively higher rating or cheaper cost?
Plot for Votes V/S Rate:¶
trace = go.Scatter(x = zomato['Rating'], y = zomato['Total_Votes'], textposition = 'top center', text = zomato['Name'], mode = 'markers')
data1 = [trace]
layout = go.Layout(title = 'Votes v/s Rating', xaxis = dict(title = 'Average Rate'), yaxis = dict(title = 'Total no. of Votes'))
fig = go.Figure(data = data1, layout = layout)
py.iplot(fig);
We see that there is a positive correlation between no of votes and the rating. As the number of votes are increasing the rating of the restaunrant also is going up. Maximum reataunrants have received a rating between 4 and 5. Byg Brewski, Truffles and Absolute Barbeque have received the maximum votes and the highest rating among all the restaurants.
Lets see the relationship between Rating and Cost of meal at a restaurant
trace = go.Scatter(x = zomato['Rating'], y = zomato['Approx_Cost'], mode = 'markers', text = zomato['Name'])
data = [trace]
layout = go.Layout(title = 'Rating v/s Cost of meal at he restaurant', xaxis = dict(title = 'Rate'), yaxis = dict(title = 'Approx_Cost'))
fig = go.Figure(data = data, layout = layout)
py.iplot(fig);
Are Cost of Restaurants in a particular Locatility costlier than other?
plt.figure(figsize=(15,5))
sns.barplot(x = 'Locality', y = 'Approx_Cost', data = zly);
plt.xticks(rotation = 60);
plt.xlabel('Locality', size = 15)
plt.ylabel('Approximate Cost', size = 15)
plt.title('Average Cost of meal at restaurants by Locality', size = 15);
Here we see that Church Street, Brigade Road, Lavelle Road are among the costiest places for a meal and Banashankari has the cheapest meals.
zomato.groupby(['Listed_Type'])['Rating'].agg(np.mean).reset_index(). set_index('Listed_Type').sort_values(['Rating'], ascending = False).plot.bar();
plt.xticks(rotation = 60);
plt.ylabel('Rating');
Pubs and bars and drinks are highest rated category of restaurants.
sns.catplot(x = 'Online_Order', y = 'Total_Votes', kind="violin", data = zomato)
plt.title("Plot of Total votes v/s Online order", size = 15) ;
Now lets take a look at the ratings of the restaurants with maximum number of branches in the city and see of there is any relation:
fig = px.scatter(top_brands, x="Avg_Rating", y="#OfBranches", trendline="ols")
fig.show()
Seems like there is a relation between rating and number of branches opened for each brand.
THE END!¶